iiRecord
Agentic AI Atlas · BBQ (Bias Benchmark for QA)
benchmark:bias-bencha5c.ai
II.
Benchmark JSON

benchmark:bias-bench

Structured · live

BBQ (Bias Benchmark for QA) json

Inspect the normalized record payload exactly as the atlas UI reads it.

File · benchmarks/benchmarks/benchmarks-knowledge.yamlCluster · benchmarks
Record JSON
{
  "id": "benchmark:bias-bench",
  "_kind": "Benchmark",
  "_file": "benchmarks/benchmarks/benchmarks-knowledge.yaml",
  "_cluster": "benchmarks",
  "attributes": {
    "displayName": "BBQ (Bias Benchmark for QA)",
    "homepageUrl": "https://github.com/nyu-mll/BBQ",
    "kind": "model-only",
    "targetsKind": "ModelVersion",
    "description": "BBQ (Parrish et al., NYU 2022) is a bias benchmark for question\nanswering covering nine social dimensions (age, disability, gender,\nnationality, physical appearance, race/ethnicity, religion,\nsocioeconomic status, sexual orientation), measuring stereotype\nbias under both ambiguous and disambiguated contexts.\n"
  },
  "outgoingEdges": [
    {
      "from": "benchmark:bias-bench",
      "to": "skill-area:safety-redteaming",
      "kind": "covers",
      "attributes": {}
    },
    {
      "from": "benchmark:bias-bench",
      "to": "domain:social-sciences",
      "kind": "applies_to",
      "attributes": {}
    }
  ],
  "incomingEdges": []
}